| name | clinical-reasoning-llm-hepatocellular-carcinoma-risk-stratification |
| description | HCC-STAR: clinically aligned LLM for hepatocellular carcinoma staging, treatment, and prognosis. Reads EMR narratives, outputs risk stratification, guideline-consistent treatments with rationales, and survival estimates. Outperforms GPT-5 and Gemini-2.5 Pro. Activation: clinical-reasoning LLM, hepatocellular carcinoma, risk stratification, treatment guidance, EMR. |
| metadata | {"arxiv_id":"2607.08602","published":"2026-07-09","authors":"Peng Cui, Jitao Wang, Siyan Xue, Yao Huang, Haoming Xia","tags":["clinical-reasoning-llm","hepatocellular-carcinoma","risk-stratification","treatment-guidance","electronic-medical-records"]} |
Towards Precision Therapy in Hepatocellular Carcinoma: A Clinical-Reasoning LLM for Risk Stratification and Treatment Guidance
Overview
HCC-STAR (Hepatocellular Carcinoma Staging, Treatment And pRognosis) is a clinically aligned large language model that reads routine EMR narratives and jointly outputs risk score-based staging, ranked guideline-consistent treatments with evidence-based rationales, and individualized survival estimates for hepatocellular carcinoma.
Key Innovations
EMR Narrative Processing
- Reads routine electronic medical record narratives directly
- Extracts clinical context missed by coarse staging systems
- Addresses within-stage heterogeneity in HCC
Knowledge-Aligned Reasoning Framework
- Optimized with step-verifiable composite reward
- Moves beyond text-level memorization of clinical guidelines
- Generates evidence-based rationales for treatment recommendations
Multi-Center Validation
- Trained on ~30,000 HCC cases from SEER, expanded into EMR-style narratives
- Validated on 6,668 patients from 12 hospitals in China
- Outperforms GPT-5 and Gemini-2.5 Pro in treatment recommendation
- Blinded hepatobiliary specialists rate reasoning as trustworthy
Survival Impact
- Median survival of 51 months under HCC-STAR recommendations
- vs. 29 months (BCLC) and 32 months (CNLC) under standard guidelines
- Helps physicians make more accurate decisions faster
Methodology
- Data Curation: 30,000 HCC cases from SEER expanded into EMR narratives via clinician-validated augmentation
- Training: Knowledge-aligned reasoning with step-verifiable composite reward
- Multi-Center Validation: 6,668 patients across 12 hospitals
- Clinician-Centric Evaluation: Blinded specialist ratings on reasoning and evidence quality
Implications
- LLM-based clinical decision support can meaningfully improve patient outcomes
- EMR narrative understanding captures context missed by staging systems
- Step-verifiable rewards enable aligned clinical reasoning
- Outperforming frontier general LLMs shows value of domain-specific training
Pitfalls
- EMR narrative augmentation may not capture all clinical nuances
- Multi-center validation is China-specific — generalization to other populations needs testing
- LLM recommendations should supplement, not replace, clinician judgment
- Survival analysis is hypothetical — prospective validation needed
Activation Keywords
clinical-reasoning LLM, hepatocellular carcinoma, HCC-STAR, risk stratification, treatment guidance, EMR processing, precision therapy, survival estimation, clinical decision support
Paper Reference
arXiv:2607.08602 - "Towards Precision Therapy in Hepatocellular Carcinoma: A Clinical-Reasoning LLM" (Jul 2026)